CREDIT RISK ANALYSIS: AN ASSESSMENT OF THE PERFORMANCE OF SIX MACHINE LEARNING TECHNIQUES IN CREDIT SCORING MODELLING

Abstract

This study checked the credit risk analysis domain, concentrating on assessing the efficacy of six distinct credit scoring methodologies: linear discriminant analysis, logistic regression, artificial neural networks, support vector machine, decision tree and, K-nearest neighbour on microcredit applicant’s data. Two performance metrics were used: Area under the receiver operative characteristic curve and, Precision. The results obtained from the experimentation phase reveal distinct performance levels for each technique. Specifically, K-nearest neighbour and artificial neural networks showcase exceptional prowess, yielding an AUC of 0.9833 and 0.9062 and, an impressive precision score of 0.8065 and 1 respectively. In contrast, logistic regression and support vector machine demonstrate a good performance with an area under the curve value of 0.8537 and 0.8532 respectively, on precision metric score, support vector machine showed impressive high performance while logistic regression performed poorly. Linear discriminant analysis and Decision tree exhibit comparatively moderate accuracy scores and achieved an AUC of 0.8494318 and 0.7524 respectively. Thus, we underscore the potential of K-nearest neighbour and Artificial neural networks as a superior method for credit risk analysis, supported by robust performance metrics. Although, all techniques achieve significantly good discriminative power and good precision. The findings advocate for the adoption of modern techniques in credit scoring modelling, positioning K-nearest neighbour and Artificial neural networks as a valuable tool in financial institutions’ risk assessment processes.